High-precision prediction method and system for stamping springback of high-strength plate part of new energy automobile

Abstract

The invention relates to the technical field of new energy automobile part stamping prediction, and discloses a new energy automobile high-strength plate part stamping springback high-precision prediction method and system, and the method comprises the steps: obtaining a dislocation density field and grain orientation distribution data of a high-strength steel plate; constructing a plastic flow equation, and calculating a curvature tensor; establishing a statistical distribution model of material parameters; thermal-mechanical coupling solution is realized through an operator splitting algorithm; integrating a thermal-mechanical coupling solving result into a digital twin system; carrying out adaptive adjustment and optimization on the plastic flow equation and the statistical distribution model; the system comprises a data acquisition module, a modeling module, a solving module, a digital twin module and an optimization module. According to the method, the synchrotron radiation X-ray diffraction and electron back scattering diffraction cooperative detection technology is combined with a plastic flow equation construction method based on differential geometry, so that accurate correlation between the material microstructure and the macroscopic mechanical behavior is realized, and the springback prediction accuracy is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of stamping prediction technology for new energy vehicle parts, specifically to a method and system for high-precision prediction of springback during stamping of high-strength steel plate parts for new energy vehicles. Background Technology

[0002] The demand for lightweight materials in new energy vehicles is experiencing explosive growth. The higher the material strength, the more difficult it is to control springback during stamping. Production line data from a major automaker shows that DP1180 steel can achieve a springback of up to 45% of its thickness, more than three times higher than traditional steel.

[0003] The existing high-precision prediction technology for stamping springback of new energy vehicle parts, according to data released by Toyota, requires an average of 5-6 trial runs for each new mold; the second is numerical simulation, with software like AutoForm being the most widely used; and the third is optical scanning compensation, typically GOM's ATOS system.

[0004] However, existing high-precision prediction technologies for springback in stamping of new energy vehicle parts only have a resolution of 50μm using ordinary X-ray diffractometers, resulting in a dislocation density calculation error of ±40%. Furthermore, the simulation and actual springback angles deviate significantly, requiring engineers to manually correct them using empirical formulas. Therefore, this invention provides a high-precision prediction method and system for springback in stamping of high-strength steel sheet parts for new energy vehicles, addressing the shortcomings of existing technologies. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides a method and system for highly accurate prediction of stamping springback in high-strength steel plate parts for new energy vehicles, solving the problems of low material characterization accuracy, idealized constitutive models, low computational efficiency, and outdated offline parameter tuning in existing technologies.

[0006] To achieve the above objectives, the present invention provides a high-precision prediction method for the springback of high-strength sheet metal parts for new energy vehicles, comprising the following steps:

[0007] Data on dislocation density field and grain orientation distribution of high-strength steel plates were obtained by synchrotron X-ray diffraction and electron backscatter diffraction.

[0008] Based on the acquired dislocation density field data, a differential geometry method considering dislocation-induced torsion is used to construct the plastic flow equation and calculate the curvature tensor.

[0009] Based on the acquired grain orientation distribution data, a statistical distribution model of material parameters is established using the random field expansion method, and thermodynamic constraints are introduced.

[0010] Based on the plastic flow equation and statistical distribution model, a thermo-mechanical coupling solution is achieved through an operator splitting algorithm, and GPU is used to accelerate tensor operations.

[0011] The results of the thermo-coupling solution are integrated into the digital twin system to output the mold compensation amount in real time;

[0012] Based on real-time monitoring data during the stamping process, the plastic flow equation and statistical distribution model are adaptively adjusted and optimized.

[0013] Preferably, the scanning parameters of the synchrotron X-ray diffraction include:

[0014] The energy of the X-rays is 30-40 keV;

[0015] The scan step size is 1-5 μm;

[0016] The single-point exposure time is 0.3-1.0 seconds.

[0017] Preferably, the construction of the plastic flow equation using a differential geometry method that considers dislocation-induced torsion includes the following steps:

[0018] Calculate the torsion tensor components based on dislocation density field data;

[0019] The torsion tensor is introduced into the calculation of the connection coefficient in the plastic flow equation;

[0020] The curvature tensor is used to characterize the geometric constraint effect of dislocations on plastic flow.

[0021] Preferably, the step of establishing a statistical distribution model of material parameters using the random field expansion method includes the following steps:

[0022] Spatial distribution data of material parameters were obtained through microindentation testing.

[0023] Perform covariance analysis on spatially distributed data to solve for the characteristic function and eigenvalues;

[0024] Select the top N eigenfunctions whose eigenvalues ​​are greater than a set threshold and perform random field expansion.

[0025] Preferably, the covariance analysis uses a Matérn-type covariance kernel function, the expression of which is:

[0026]

[0027] Where x1 and x2 represent the coordinates of any two points in space; For variance; l c The relevant length is denoted by ||x1-x2||, which represents the Euclidean distance between the two points.

[0028] Preferably, the thermal coupling solution achieved through the operator splitting algorithm includes the following steps:

[0029] Perform mechanical step calculations and update the test stress tensor;

[0030] The Moreau-Yosida regularization method was used to calculate the projection step.

[0031] The microstructure parameters are updated using the dislocation density transport equation.

[0032] Preferably, integrating the results of the thermo-coupling solution into the digital twin system includes the following steps:

[0033] Process data is collected in real time using sensors on the stamping machine, with a sampling frequency of no less than 1kHz.

[0034] The stream processing engine is used to extract features from the collected data and generate a prediction input vector.

[0035] The input vector is fed into the prediction model optimized by TensorRT, and the output is the mold compensation amount.

[0036] Preferably, the formula for calculating the mold compensation amount is:

[0037] ΔS=-α·J T ·U springback ;

[0038] Where ΔS represents the compensation amount of the mold surface; α is the compensation coefficient, ranging from 0.6 to 1.2; J is the process sensitivity matrix; U springback This is the rebound displacement field.

[0039] Preferably, the adaptive adjustment optimization includes the following steps:

[0040] The pressure sensor monitors the fluctuation data of the blank holder force in real time during the stamping process;

[0041] When the rebound prediction residual is detected to exceed 0.2mm, the parameter update process is triggered;

[0042] The hardening coefficient in the plastic flow equation and the covariance parameter in the random field model are adjusted using the stochastic gradient descent algorithm.

[0043] A high-precision prediction system for stamping springback of high-strength sheet metal parts for new energy vehicles is also provided, including:

[0044] The data acquisition module is used to acquire dislocation density field and grain orientation distribution data of high-strength steel plates;

[0045] The modeling module is used to construct plastic flow equations and statistical distribution models of material parameters;

[0046] The solver module is used to implement thermo-mechanical coupling solutions;

[0047] The digital twin module is used to output mold compensation amounts in real time;

[0048] The optimization module is used to adaptively adjust model parameters.

[0049] This invention provides a method and system for highly accurate prediction of springback during stamping of high-strength steel sheet parts for new energy vehicles. It offers the following advantages:

[0050] 1. This invention utilizes a combined detection technique of synchrotron X-ray diffraction and electron backscatter diffraction, along with a method for constructing plastic flow equations based on differential geometry, to achieve a precise correlation between the microstructure and macroscopic mechanical behavior of materials. Compared to existing technologies that employ the assumption of uniform dislocation density in their modeling, this invention effectively solves the technical challenge of accurately characterizing local deformation gradients using traditional methods, significantly improving the accuracy of springback prediction.

[0051] 2. The operator splitting algorithm based on graphics processor acceleration designed in this invention, combined with optimization by a deep learning inference engine, significantly shortens the computation time of a single complete simulation. Compared with the computation mode of traditional implicit algorithms, this invention significantly improves the solution efficiency of large-scale engineering problems through dynamic batch processing and layer fusion technology.

[0052] 3. The online parameter adjustment mechanism established in this invention employs a constrained optimization algorithm to achieve dynamic optimization of key model variables. This system can automatically identify deviations caused by process fluctuations and precisely adjust the compensation amount, significantly reducing the number of mold adjustments compared to traditional trial-and-error methods, and overcoming the technical shortcomings of the lag in manual parameter adjustment.

[0053] 4. The virtual-real mapping system constructed in this invention achieves end-to-end connectivity from data acquisition to compensation execution through a standardized data interface. Compared to existing digital twin systems that only have monitoring functions, this invention forms a complete process optimization closed loop, effectively overcoming the technical limitations of insufficient control capabilities in traditional systems. Attached Figure Description

[0054] Figure 1 This is a flowchart of the method steps of the present invention;

[0055] Figure 2 This is a system architecture diagram of the present invention. Detailed Implementation

[0056] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0057] Please see the appendix Figure 1This invention provides a method for highly accurate prediction of springback during stamping of high-strength steel sheet parts for new energy vehicles, comprising the following steps:

[0058] S1. Obtain dislocation density field and grain orientation distribution data of high-strength steel plates by synchrotron X-ray diffraction and electron backscatter diffraction.

[0059] S2. Based on the acquired dislocation density field data, a plastic flow equation is constructed using a differential geometry method that considers dislocation-induced torsion, and the curvature tensor is calculated.

[0060] S3. Based on the acquired grain orientation distribution data, a statistical distribution model of material parameters is established using the random field expansion method, and thermodynamic constraints are introduced.

[0061] S4. Based on the plastic flow equation and statistical distribution model, a thermo-mechanical coupling solution is achieved through the operator splitting algorithm, and GPU is used to accelerate tensor operations;

[0062] S5. Integrate the results of the thermo-coupling solution into the digital twin system and output the mold compensation amount in real time;

[0063] S6. Based on real-time monitoring data during the stamping process, the plastic flow equation and statistical distribution model are adaptively adjusted and optimized.

[0064] For step S1, in this embodiment, by combining synchrotron X-ray diffraction and electron backscatter diffraction techniques, multi-scale material characterization from macro to micro is achieved, ensuring that the obtained dislocation density field and grain orientation distribution data have sufficient spatial resolution and statistical reliability.

[0065] Generally, the high-strength steel plate sample to be tested is first subjected to surface treatment. Specifically, electropolishing is performed at 20V for 90 seconds to obtain a surface condition that meets the requirements of EBSD testing. As an option, for advanced high-strength steel with a high silicon content, argon ion polishing can be used for additional surface finishing.

[0066] In one possible implementation, synchrotron X-ray diffraction tests are performed on a third-generation synchrotron radiation source. The test parameters are set as follows: X-ray energy 35 keV, beam size 0.5 × 0.5 μm. 2 The step size is 2μm, and the exposure time is 0.5 seconds per point. In particular, for substrates thicker than 2mm, a double-sided scanning strategy is required to eliminate information loss in the thickness direction.

[0067] Specifically, the dislocation density ρ is calculated using a modified Williamson-Hall method:

[0068]

[0069] Where ρ represents the dislocation density; K is the geometric factor (value 1.67); ω is the measured lattice distortion angle; b is the Burgers vector mode; and A is the integral area of ​​the diffraction peak.

[0070] In some embodiments, EBSD detection is performed using a field emission scanning electron microscope (FET) in conjunction with a high-speed detector. Detection conditions include: accelerating voltage 20 kV, beam current 6 nA, step size 0.5 μm, and acquisition speed 300 points / second. It is particularly important to note that for materials with significant texture, a helical scanning path should be used to avoid orientation sampling bias.

[0071] Specifically, the grain orientation distribution function (ODF) is calculated using a series expansion method:

[0072]

[0073] Where f(g) is the orientation distribution function; g represents the orientation matrix; The expansion factor; is a generalized spherical harmonic function; the maximum value of l is 22 to ensure sufficient angular resolution; m is the secondary exponent of the series expansion.

[0074] In one possible implementation, data preprocessing includes the following key steps: first, background subtraction and peak correction are performed on the original diffraction pattern; then, a Gaussian mixture model is used to deconvolve the overlapping diffraction peaks; finally, the spatial distribution of the dislocation density field is reconstructed using a three-dimensional inversion algorithm. Specifically, for regions with significant gradient changes, the sampling density should be increased to 1 μm / point.

[0075] In some embodiments, strict quality control standards are set to ensure data reliability: the coefficient of variation of the full width at half maximum (FWHM) of the diffraction peaks does not exceed 15%, the EBSD Kikuchi band contrast threshold is set to 0.8, and the orientation difference angle statistical error is controlled within 0.5°. When data anomalies are detected, the system automatically triggers a local retesting process.

[0076] Specifically, the final output data package contains the following core information: the three-dimensional dislocation density distribution field ρ(x,y,z), the grain orientation matrix g, the orientation distribution function f(g), and related quality assessment metrics. This data is stored in HDF5 format and includes a complete metadata description.

[0077] For step S2, in this embodiment, a differential geometry method considering dislocation-induced torsion is used to construct the plastic flow equation and calculate the curvature tensor. By establishing a strict mathematical relationship between the microscopic dislocation structure and the macroscopic plastic deformation behavior, and by introducing the torsion tensor to characterize the geometric constraint effect of dislocations on the plastic flow of the material, the nonlinear deformation characteristics of high-strength steel plates can be described more accurately.

[0078] Specifically, based on the dislocation density field data obtained in step S1, the torsion tensor components are first calculated. Torsion tensor With dislocation density α li The relationship is:

[0079]

[0080] in, is the torsion tensor; κ is the material constant (0.42±0.03 for DP steel); ∈ jkl The symbol for Levi-Civita; α li Dislocation density tensor components (unit: m) -2 ).

[0081] In some embodiments, the torsion tensor is calculated using a discretization method, dividing the detection region into finite volume cells, with the dislocation density within each cell being a local average. For regions with significant gradient changes, the cell size is reduced to 1 / 4 of its original value to improve computational accuracy.

[0082] In one possible implementation, the plastic flow equation takes the form:

[0083]

[0084] in, Plastic strain rate; Φ is the equivalent plastic strain rate; Φ is the yield potential function; β ij Let be the dislocation mobility tensor; σ is the affine connection coefficient; kl For stress tensor; It is a partial derivative.

[0085] In general, the curvature tensor Calculated using the following formula:

[0086]

[0087] in, Characterizes the degree of local geometric distortion of the material at point (q, μ, ν); θ u θ v These represent the differential 1-form in the local coordinate system; These represent the components of the affine connection coefficient; denoted as the modified connection coefficient components; λ is the summation index.

[0088] Specifically, the following points should be noted during the calculation process:

[0089] The partial derivative terms were calculated using the central difference method with a step size of 0.1 mm.

[0090] For boundary regions, the mirror extension method is used to handle the data missing problem;

[0091] The dimension of the curvature tensor is m -2 Its physical meaning characterizes the degree of local geometric distortion caused by dislocation.

[0092] As an alternative, GPU acceleration can be used to compute the curvature tensor. In practice, the computational task is broken down into thread blocks, each responsible for computing a finite-volume unit. Global memory access employs a merged access mode, achieving a memory bandwidth utilization rate of over 80%.

[0093] In some embodiments, the curvature tensor is regularized to improve computational stability:

[0094]

[0095] in, R represents the components of the original curvature tensor; R0 is the reference curvature used to prevent numerical overflow; R is the Frobenius norm of the curvature tensor.

[0096] For step S3, in this embodiment, a statistical distribution model of the material parameters is established using the random field expansion method, and thermodynamic constraints are introduced. This step establishes a probabilistic correlation between the spatial variability of the material's microstructure and its macroscopic mechanical response, providing crucial constitutive relation inputs for subsequent thermo-mechanical coupling solutions.

[0097] Specifically, the obtained grain orientation data is first preprocessed. In one possible implementation, a moving average method is used to eliminate test noise, with the window size set to 3×3 measurement points. The preprocessed orientation data is then used to construct a random field model of the material parameters.

[0098] In some embodiments, the random field expansion employs a truncated Karhunen-Loève series:

[0099]

[0100] in, λ represents the mean field of material parameters. k The k-th eigenvalue of the covariance operator; φ k (x) is the k-th order characteristic function of the covariance operator; ξ k (ω) is an independent standard normal random variable.

[0101] As an alternative, the covariance function employs a modified Matérn model:

[0102]

[0103] Where x1 and x2 represent the coordinates of any two points in space; For variance; l c The relevant length is denoted by ||x1-x2||, which represents the Euclidean distance between the two points.

[0104] In some embodiments, the following verification criteria are set: random field reconstruction error ≤ 8%, thermodynamic constraint satisfaction rate ≥ 95%, and correlation coefficient between parameter spatial distribution and EBSD detection results ≥ 0.85;

[0105] The final output random field model includes the following elements: characteristic functions and eigenvalues, random variable distribution parameters, thermodynamic constraint parameters, and validation index report.

[0106] One possible implementation involves establishing a model database and implementing version control for easy traceability and reuse. Data is stored in HDF5 format, with each file approximately 50MB in size, containing complete metadata descriptions and version information.

[0107] For step S4, in this embodiment, the thermo-coupling problem is solved using an operator splitting algorithm, and tensor operations are accelerated using a GPU. This step combines differential geometry theory with modern numerical calculation methods, effectively solving the problem of thermo-coupling nonlinearity in the stamping process of high-strength steel plates.

[0108] Generally, the solution process consists of three computational stages. Specifically, the first step is a machine step calculation to update the test stress tensor. In one possible implementation, an explicit time integration scheme is used, with the time step automatically adjusted according to the Courant condition.

[0109] As an alternative, the NVIDIA CUDA architecture is used for parallel computing optimization. Specifically, the computation domain is divided into 256×256 thread blocks, with each thread handling computation tasks across 4 grid points. Global memory access adopts a merged access mode, maintaining memory bandwidth utilization above 85%.

[0110] In one possible implementation, the following stability conditions are set: plastic strain increment ≤ 0.002, temperature change rate ≤ 50℃ / s, and stress oscillation amplitude < 5MPa.

[0111] When instability is detected, a step size halving mechanism is automatically triggered. In some embodiments, Aitken acceleration technology is used to improve convergence, and the dynamic relaxation factor is controlled within the range of 0.6-1.4.

[0112] Specifically, the heat conduction equation and the mechanical equation are solved using an alternating iterative strategy. Generally, each mechanical step corresponds to 2-3 heat conduction steps. In one possible implementation, the heat flux term is calculated using an upwind difference scheme to ensure numerical stability.

[0113] In some embodiments, the following data specifications are set: stress tensors are stored in FP32 format, temperature fields are stored using FP16 compression, and calculation results are automatically saved as checkpoints every 10 seconds.

[0114] As an alternative, RDMA technology is used to achieve data transfer between the CPU and GPU, with latency controlled within 50μs. For large-scale computing, the NCCL communication library between multiple GPUs is enabled for data exchange.

[0115] Specifically, the performance optimization measures implement the following optimization strategies: high-frequency access variables are stored in shared memory, the transient function uses fast approximation calculation, and the kernel function startup configuration is adjusted to 128 threads / block.

[0116] One possible implementation involves setting up a dynamic load balancing mechanism to automatically adjust task allocation based on the computational load of each GPU. In some embodiments, CUDAGraph technology is used to optimize kernel function call sequences and reduce API overhead.

[0117] Generally, the following verification standards are set for the calculation results: energy conservation error <1%, momentum balance residual <0.5 N / mm. 2 Thermodynamic constraint satisfaction rate ≥ 98%.

[0118] If the verification fails, the calculation restart process is automatically triggered, and the calculation is recalculated from the most recent checkpoint. As an option, a mixed-precision calculation mode is enabled, with critical variables using FP64 precision and other variables using FP32 precision.

[0119] For step S5, in this embodiment, a digital bridge connecting numerical simulation and physical production is constructed, and online optimization and adjustment of stamping process parameters are realized through efficient data pipelines and intelligent decision-making mechanisms.

[0120] Typically, a layered distributed architecture is adopted. Specifically, it consists of a three-tiered control system comprising edge computing nodes, a cloud analytics platform, and execution terminals. In one possible implementation, the edge nodes utilize NVIDIA Jetson AGX Orin modules equipped with 16GB of video memory.

[0121] In some embodiments, a multi-level buffered data transmission channel is established:

[0122] Sensor layer: 1kHz sampling frequency;

[0123] Edge layer: 100ms time window;

[0124] Cloud-based: 1-second aggregation cycle.

[0125] As an alternative, Apache Kafka is used for data stream management, with the number of topic partitions dynamically adjusted based on the number of stamping stations. Message serialization uses Protocol Buffers format, with each compressed message being approximately 2KB.

[0126] Specifically, the formula for calculating the mold compensation amount is:

[0127] ΔS=-α·J T ·U springback ;

[0128] Where ΔS represents the compensation amount of the mold surface; α is the compensation coefficient, ranging from 0.6 to 1.2; J is the process sensitivity matrix; U springback For the rebound displacement field

[0129] In one possible implementation, TensorRT is used to optimize the prediction model:

[0130] FP16 precision inference;

[0131] Dynamic batch processing (1-8 batches);

[0132] Layer fusion technology reduces computational overhead.

[0133] In some embodiments, a hot-update mechanism is set up to automatically load a new model version when a prediction deviation > 0.2 mm is detected. The update process ensures uninterrupted service with a switching latency of < 50 ms.

[0134] As an alternative, compensatory execution control employs a three-phase execution strategy:

[0135] Virtual verification (digital twin environment);

[0136] Small-batch trial production (3-5 pieces);

[0137] Full implementation.

[0138] Specifically, the positioning accuracy of the servo drive compensation mechanism is controlled within ±0.01mm, and the response time is <100ms. In one possible implementation, a force-position hybrid control mode is set to prevent overshoot.

[0139] In some embodiments, a multi-level alarm system is established: primary warning: deviation 0.15-0.2mm, intermediate alarm: deviation 0.2-0.3mm, emergency shutdown: deviation >0.5mm.

[0140] Under normal circumstances, the root cause analysis process is automatically initiated after an alarm is triggered, and the analysis time is controlled within 5 minutes. As an alternative, fault tree analysis is used to locate the root cause of the problem.

[0141] Specifically, the data persistence management is set to a rolling storage strategy: real-time data is retained for 7 days, feature data is retained for 30 days, and model snapshots are retained for 180 days.

[0142] In one possible implementation, a time-series database is used to store sensor data, achieving a compression ratio of 10:1. Metadata management employs blockchain technology to ensure immutability.

[0143] In some embodiments, a three-dimensional visualization monitoring interface is developed to display in real time: compensation distribution cloud map, key point displacement curve, and equipment status indicator lights.

[0144] As an option, it supports VR / AR multi-mode interaction with operation latency controlled within 20ms. Alarm information uses multiple prompts of sound and light to ensure timely response.

[0145] For step S6, in this embodiment, the real-time monitoring data for data acquisition and preprocessing includes: pressure sensor signal (range 0-2000kN), displacement measurement value (accuracy ±0.01mm), and temperature reading (sampling rate 10Hz).

[0146] In one possible implementation, a sliding window filter is used to process the raw data, with the window width set to 5 stamping cycles. In some embodiments, an outlier removal mechanism is set, with the removal criterion being the ±3σ principle.

[0147] As an option, parameter update trigger conditions can be set with multi-level trigger thresholds:

[0148] Warning level: Rebound deviation 0.15-0.2mm;

[0149] Updated grade: Springback deviation > 0.2mm;

[0150] Emergency level: 3 consecutive deviations > 0.3mm.

[0151] Specifically, the triggering mechanism employs a voting system, requiring at least three independent sensors to simultaneously trigger an alarm before confirmation. In one possible implementation, a minimum update interval of 30 minutes is set to prevent frequent adjustments.

[0152] In some embodiments, parameter updates employ a modified stochastic gradient descent method:

[0153]

[0154] Where, θ t A vector of key parameters describing the mechanical and thermal properties of materials; η t The adaptive learning rate decays over time; This is the partial derivative of the loss function with respect to the parameter vector.

[0155] Generally, the adjustment range of model parameters is set with the following limitations:

[0156] Hardening coefficient: ±15% of initial value;

[0157] Covariance parameter: ±20% of initial value;

[0158] Thermodynamic coefficient: ±10% of the initial value.

[0159] As an alternative, a constrained optimization algorithm is employed to ensure that parameter adjustments do not exceed physically reasonable limits. Specifically, a convex hull boundary of the feasible region of parameters is constructed, and a projection check is performed in each iteration.

[0160] In one possible implementation, the verification and deployment process sets up a three-level verification mechanism: digital twin simulation verification, trial molding verification (3-5 pieces), and small-batch production verification (20-30 pieces).

[0161] In some embodiments, the verification criteria are: rebound prediction error < 0.15 mm, stability index > 0.85, and convergence speed < 50 iterations.

[0162] Specifically, historical data management establishes a parameter update log database, which records the following:

[0163] Adjust timestamp;

[0164] Parameter change;

[0165] Performance evaluation indicators;

[0166] Staff information.

[0167] One possible implementation uses blockchain technology to ensure the logs are immutable. Data retention is set to 3 years, and fast retrieval by time range is supported.

[0168] As an alternative, the exception handling mechanism includes the following safeguards:

[0169] Limitation on the range of parameter adjustment per transaction;

[0170] Limit on the number of consecutive adjustments;

[0171] Rollback mechanism (automatically restores to the previous stable version).

[0172] In some embodiments, when system oscillation is detected (three consecutive positive and negative parameter adjustments), an expert consultation mode is automatically triggered, and the automatic adjustment process is paused.

[0173] Specifically, the performance monitoring interface features a dedicated monitoring panel that displays: historical parameter adjustment curves, model prediction error trends, system stability indicators, and real-time alarm status.

[0174] One possible implementation supports pushing important alarm information to mobile devices, with response latency controlled to within 1 minute. Historical data can be used to generate PDF reports for easy quality traceability.

[0175] The high-precision prediction system for the springback of high-strength steel sheet parts for new energy vehicles described below can be used as a reference to the high-precision prediction method for the springback of high-strength steel sheet parts for new energy vehicles described above.

[0176] Please see the appendix Figure 2 This invention also provides a high-precision prediction system for the springback of high-strength steel sheet parts for new energy vehicles, comprising:

[0177] The data acquisition module is used to acquire dislocation density field and grain orientation distribution data of high-strength steel plates. It employs a combined synchrotron X-ray diffractometer and electron backscatter diffraction system, equipped with a high-precision positioning platform (repeatability ±1μm) and temperature compensation device. The module has a built-in data preprocessing unit that performs real-time noise filtering (Gaussian filtering, σ = 0.8) and outlier removal (3σ criterion), outputting standardized dislocation density field data (HDF5 format) and grain orientation matrix (Bunge notation).

[0178] The modeling module is used to construct plastic flow equations and statistical distribution models of material parameters; it integrates a differential geometry calculation engine (supporting CUDA acceleration) and a random field analysis toolkit. Specifically, it includes: a plastic flow equation builder employing an improved Cartan connection algorithm (κ = 0.42 ± 0.03); and a material parameter statistical model generator, implementing Karhunen-Loève expansion (truncated order 5-7) and Matérn covariance kernel function fitting (lc = 2.5 mm). The module outputs encrypted two-dimensional flow model files (.bsm format).

[0179] The solution module is used to implement thermo-coupling solutions; it deploys a multiphysics coupled solver, whose core components include: an operator splitting computation unit (mechanical step-projection step-transport step loop); a thermo-coupling iterative controller (maximum 50 iterations, residual threshold 1e-6); and a GPU-accelerated computation framework (optimized based on the NVIDIA Ampere architecture). The solution accuracy is controlled within: stress error ±3MPa, temperature error ±2℃.

[0180] The digital twin module is used to output mold compensation amounts in real time; it constructs a virtual-real mapping system, including: a real-time data pipeline (Apache Kafka, throughput ≥10MB / s); a compensation amount calculation engine (TensorRT optimized, latency <50ms); and a 3D visualization interface (Unity3D rendering). It outputs mold surface correction instructions (G-code format) with an execution cycle ≤100ms.

[0181] The optimization module is used to adaptively adjust model parameters. It implements an adaptive parameter adjustment mechanism and includes: an online monitoring unit (sampling rate 1kHz); an incremental learning algorithm library (SGD, Adam, etc.); and a security protection system (parameter change constraint ±20%). Update trigger conditions: rebound deviation > 0.2mm or three consecutive prediction errors exceeding the limit. Historical data is stored using blockchain (HyperledgerFabric framework).

[0182] The system in this embodiment can be used to execute the above method embodiments, and its principle and technical effect are similar, so they will not be described again here.

[0183] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A high-precision prediction method for stamping springback of high-strength sheet metal parts for new energy vehicles, characterized in that, Includes the following steps: Data on dislocation density field and grain orientation distribution of high-strength steel plates were obtained by synchrotron X-ray diffraction and electron backscatter diffraction. Based on the acquired dislocation density field data, a differential geometry method considering dislocation-induced torsion is used to construct the plastic flow equation and calculate the curvature tensor. Based on the acquired grain orientation distribution data, a statistical distribution model of material parameters is established using the random field expansion method, and thermodynamic constraints are introduced. Based on the plastic flow equation and statistical distribution model, a thermo-mechanical coupling solution is achieved through an operator splitting algorithm, and GPU is used to accelerate tensor operations. The results of the thermo-coupling solution are integrated into the digital twin system to output the mold compensation amount in real time; Based on real-time monitoring data during the stamping process, the plastic flow equation and statistical distribution model are adaptively adjusted and optimized.

2. The method for high-precision prediction of stamping springback of high-strength steel plate parts for new energy vehicles according to claim 1, characterized in that, The scanning parameters of the synchrotron X-ray diffraction include: The energy of the X-rays is 30-40 keV; The scan step size is 1-5 μm; The single-point exposure time is 0.3-1.0 seconds.

3. The method for high-precision prediction of stamping springback of high-strength steel plate parts for new energy vehicles according to claim 1, characterized in that, The method of constructing the plastic flow equation using differential geometry that considers dislocation-induced torsion includes the following steps: Calculate the torsion tensor components based on dislocation density field data; The torsion tensor is introduced into the calculation of the connection coefficient in the plastic flow equation; The curvature tensor is used to characterize the geometric constraint effect of dislocations on plastic flow.

4. The method for high-precision prediction of stamping springback of high-strength steel plate parts for new energy vehicles according to claim 1, characterized in that, The establishment of a statistical distribution model for material parameters using the random field expansion method includes the following steps: Spatial distribution data of material parameters were obtained through microindentation testing. Perform covariance analysis on spatially distributed data to solve for the characteristic function and eigenvalues; Select the top N eigenfunctions whose eigenvalues ​​are greater than a set threshold and perform random field expansion.

5. The method for high-precision prediction of stamping springback of high-strength steel plate parts for new energy vehicles according to claim 4, characterized in that, The covariance analysis uses a Matérn-type covariance kernel function, the expression of which is: Where x1 and x2 represent the coordinates of any two points in space; For variance; l c The relevant length is denoted by ||x1-x2||, which represents the Euclidean distance between the two points.

6. The method for high-precision prediction of stamping springback of high-strength steel plate parts for new energy vehicles according to claim 1, characterized in that, The method of achieving thermo-coupling solution through operator splitting algorithm includes the following steps: Perform mechanical step calculations and update the test stress tensor; The Moreau-Yosida regularization method was used to calculate the projection step. The microstructure parameters are updated using the dislocation density transport equation.

7. The method for high-precision prediction of stamping springback of high-strength steel plate parts for new energy vehicles according to claim 1, characterized in that, Integrating the results of the thermo-coupling solution into the digital twin system includes the following steps: Process data is collected in real time using sensors on the stamping machine, with a sampling frequency of no less than 1kHz. The stream processing engine is used to extract features from the collected data and generate a prediction input vector. The input vector is fed into the prediction model optimized by TensorRT, and the output is the mold compensation amount.

8. The method for high-precision prediction of stamping springback of high-strength steel plate parts for new energy vehicles according to claim 7, characterized in that, The formula for calculating the mold compensation amount is: ΔS=-α·J T ·U springback ; Where ΔS represents the compensation amount of the mold surface; α is the compensation coefficient, ranging from 0.6 to 1.2; J is the process sensitivity matrix; U springback This is the rebound displacement field.

9. The method for high-precision prediction of stamping springback of high-strength steel plate parts for new energy vehicles according to claim 1, characterized in that, The adaptive adjustment and optimization includes the following steps: The pressure sensor monitors the fluctuation data of the blank holder force in real time during the stamping process; When the rebound prediction residual is detected to exceed 0.2mm, the parameter update process is triggered; The hardening coefficient in the plastic flow equation and the covariance parameter in the random field model are adjusted using the stochastic gradient descent algorithm.

10. A high-precision prediction system for the springback of stamping of high-strength steel sheet parts for new energy vehicles, applied to the high-precision prediction method for the springback of stamping of high-strength steel sheet parts for new energy vehicles as described in any one of claims 1-9, characterized in that, include: The data acquisition module is used to acquire dislocation density field and grain orientation distribution data of high-strength steel plates; The modeling module is used to construct plastic flow equations and statistical distribution models of material parameters; The solver module is used to implement thermo-mechanical coupling solutions; The digital twin module is used to output mold compensation amounts in real time; The optimization module is used to adaptively adjust model parameters.

Images